Generative adversarial network and gradient boosting decision tree model-based alloy metal material mechanical and reaming performance prediction method

Abstract

The embodiment of the invention discloses an alloy metal material mechanical and reaming performance prediction method based on a generative adversarial network and a gradient boosting decision tree model. The method comprises the following steps: establishing an original performance database of an alloy metal material; expanding the performance database of the alloy metal material by utilizing a generative adversarial network to obtain an expanded performance database, and constructing a mechanical and reaming performance database; dividing data of the mechanical and reaming performance database into training set data and test set data; dividing training set data into training data and verification data by adopting a cross validation method, performing learning training on the training data by utilizing the gradient boosting decision tree model, correcting hyper-parameters of the gradient boosting decision tree model by utilizing the verification data, and constructing the gradient boosting decision tree model; evaluating the constructed gradient boosting decision tree model by using the test set data; and utilizing the constructed gradient lifting decision tree model to predict the mechanical and reaming performance of the alloy metal material. The method is high in prediction result interpretability and high in prediction precision.

Description

Technical Field

[0001] The present invention belongs to the technical field of alloy material performance prediction, and specifically relates to a method for predicting the mechanical and hole expansion properties of alloy-based metal materials based on a generative adversarial network and a gradient boosting decision tree model. Background Art

[0002] Hole expansion performance is the hole expansion and flanging performance, which refers to the ability of metal materials to withstand deformation without cracking during hole expansion or flanging operations in the manufacturing process. Automotive parts such as suspension brackets and connectors, door anti-collision beams, body structural parts, hubs and wheels, etc., not only require high strength, but also require excellent edge quality and crack resistance during the forming process, thereby improving the overall safety and durability of the vehicle, so hole expansion and flanging performance is one of the important technical indicators of automotive parts.

[0003] In the prior art, the hole expansion experiment is mainly used to test the hole expansion and flanging performance of metal materials, that is, a hole of standard diameter is punched in the center of the metal material sample, and the hole diameter is gradually expanded by a tapered punch until obvious cracks appear at the edge of the hole, then the hole expansion operation is stopped and the final hole diameter at this time is recorded; the hole expansion rate HER is calculated by the ratio of the difference between the final hole diameter Df after hole expansion and the initial hole diameter Do to the initial hole diameter, and the formula is as follows:

[0004]

[0005] Where HER is the hole expansion ratio, D f is the final aperture, D o is the initial aperture.

[0006] However, the number of samples required to test the hole expansion and flanging performance through experimental methods is large and the error value is large; the experimental environment and processing method have a great influence on the test results of the hole expansion experiment, and the experimental method of testing the hole expansion rate will consume a lot of manpower cost, time cost and economic cost.

[0007] In recent years, with the development of computer science and data science, machine learning methods have been applied in material design and mechanical property evaluation. However, machine learning has made some progress in the prediction of metal mechanical properties, but it is mainly for single materials to predict within a narrow range of alloy composition, and it is difficult to use the powerful data mining capabilities of machine learning models. For example, Chinese patent CN112685962A discloses a material yield strength prediction method based on machine learning. The article uses the microstructure parameters, intermediate calculation parameters and corresponding yield strength of the test material as the yield strength data group; the ELM model is used to train and test all the data to obtain the final prediction model, and the prediction model is used to predict the yield strength of the material to be tested. This method makes up for the shortcomings of high material cost, time cost and calculation cost when predicting yield strength by traditional methods. Chinese patent CN112951344A discloses a method for predicting uniform elongation of materials based on machine learning. First, the first extreme learning machine model is trained and tested according to the standard data set to obtain the initial uniform elongation prediction model; then, the microstructure parameters and uniform elongation of several groups of materials to be tested are obtained in a targeted manner to obtain the real data set and the second uniform elongation prediction model; then, a virtual sample generation method combined with a particle swarm optimization algorithm is used to generate virtual samples around a part of the real samples; finally, the second uniform elongation prediction model is used to train and test all samples to obtain the final uniform elongation prediction model. However, the method disclosed in the above patent document is only applicable to the prediction of mechanical properties of a bimodal structure metal, and the amount of data is relatively limited, which is not universal. Summary of the invention

[0008] In view of this, some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, comprising the steps of:

[0009] S1. Establish an original performance database of alloy metal materials; the original performance database includes the components and contents, mechanical performance parameters and hole expansion performance parameters of alloy metal materials, which are collected through two channels: literature and experimental data;

[0010] S2. Using a generative adversarial network, the original performance database of alloy-based metal materials is expanded to obtain an expanded performance database; the original performance database and the expanded performance database constitute a mechanical and hole expansion performance database of alloy-based metal materials;

[0011] S3, using stratified sampling method, the data of mechanical and hole expansion performance database is divided into training set data and test set data;

[0012] S4, using the cross-validation method to divide the training set data into training data and validation data, using the gradient boosting decision tree model to learn and train the training data, using the validation data to calibrate the hyperparameters of the gradient boosting decision tree model, and constructing the gradient boosting decision tree model;

[0013] S5. Use the test set data to evaluate the constructed gradient boosting decision tree model to evaluate its prediction accuracy, generalization ability and prediction stability;

[0014] S6. Use the constructed gradient boosting decision tree model to predict the mechanical properties and hole expansion performance of alloy metal materials.

[0015] Further, some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, wherein the generative adversarial network is composed of a generator and a discriminator, and a method for obtaining an expanded performance database using the generative adversarial network includes:

[0016] S201, obtaining initial model parameters of the generator and the discriminator;

[0017] S202, using the original performance database data as the real database data of the discriminator, and generating random noise data as the input data of the generator;

[0018] S203, training the discriminator, specifically including: extracting a batch of real sample data from the real database data, using the generator to generate a batch of fake sample data, and calculating the loss function using the following formula:

[0019]

[0020] Among them, the loss function encourages the discriminator to classify the real sample data as real and the generated fake sample data as fake, and updates the parameters of the discriminator by minimizing the loss function of the discriminator;

[0021] S204, training generator; specifically including:

[0022] Using the fake sample data generated by the generator, the loss function of the generator is calculated using the following formula:

[0023] L G =-E z～pz(z) [logD(G(z))];

[0024] The generator minimizes the loss function so that the discriminator cannot distinguish between the generated fake sample data and the real sample data, and updates the parameters of the generator by minimizing the loss function of the generator;

[0025] S205, repeat steps S203 and S204 multiple times;

[0026] S206. Adjust the hyperparameters of the generative adversarial network according to the quality of the generated fake sample data until the quality of the generated fake sample data reaches the expected level; the hyperparameters include the learning rate, the number of training times of the generator, and the number of training times of the discriminator;

[0027] S207. Generate sample data that meets expected quality using the generator of the trained generative adversarial network to obtain an expanded performance database.

[0028] Some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy-based metal materials based on a generative adversarial network and a gradient boosting decision tree model. In step S3, a stratified sampling method is used to perform stratified sampling in a ratio of 9:1 to divide the data in the performance prediction database into training set data and test set data.

[0029] In some embodiments, the method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model is disclosed, and step S4 includes:

[0030] S401, using training data to initialize the gradient boosting decision tree model, for the mean square error loss function, the initial prediction value is the mean of the target value of the training data, and the calculation formula is:

[0031]

[0032] Among them, L is the loss function, yi is the target value, and F o(x) is the initial predicted value, γ is the true value;

[0033] S402, calculating the residual between the predicted value and the true value of each alloy metal material sample as the error of the current gradient boosting decision tree model, wherein the residual calculation formula for the mth iteration is:

[0034]

[0035] Among them, r im represents the residual of the ith alloy-based metal material sample of the mth iteration;

[0036] S403, taking the calculated residual as the target value, training a new decision tree, and fitting the residual; wherein the decision tree of the mth iteration is expressed as:

[0037] h m (x) = Train(X, r m );

[0038] Among them, X is the training data set, r m is the residual of the current alloy metal material sample at the mth iteration, h m (x) is the regression tree of round m;

[0039] S404, adding the new decision tree to the gradient boosting decision tree model, and updating the prediction value of the model; the updated gradient boosting decision tree model is:

[0040] F m (x) = F m-1 (x)+αh m (x)

[0041] Among them, F m (x) is the mth predicted value, F m-1 (x) is the m-1th prediction value, α is the learning rate, which ranges from 0 to 1; h m (x) is the regression tree of the mth iteration;

[0042] S405, repeat steps S402 to S404 until the predetermined number of iterations M is reached or the model performance requirements are met, and the final gradient boosting decision tree model is obtained, which is expressed as:

[0043]

[0044] Among them, F M (x) is the predicted value after the Mth iteration; F0(x) is the initial predicted value; α is the learning rate, which ranges from 0 to 1; h m (x) is the mth regression tree; m is the number of iterations, ranging from 1 to M.

[0045] Some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, which uses a gradient boosting decision tree model to learn training data and uses grid search and cross validation to optimize parameters, wherein the cross validation is selected to be 10 fold.

[0046] Some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, and use the mean square error MSE to evaluate the generalization ability of the gradient boosting decision tree model. The calculation formula of MSE is:

[0047]

[0048] Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio.

[0049] Some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, using the average relative error AARE and the determination coefficient R 2Evaluate the prediction accuracy of the gradient boosted decision tree model;

[0050] The calculation formula of average relative error AARE is:

[0051]

[0052] Coefficient of determination R 2 The calculation formula is:

[0053]

[0054] Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio.

[0055] In some embodiments, the prediction method for the mechanical and hole expansion properties of alloy metal materials based on the generative adversarial network and the gradient boosting decision tree model is disclosed, and the standard deviation SD is used to evaluate the stability of the gradient boosting decision tree model. The calculation formula of SD is:

[0056]

[0057] Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio, and AARE is the average relative error.

[0058] The embodiment of the present invention discloses a method for predicting the mechanical and hole expansion properties of alloy-based metal materials based on a generative adversarial network and a gradient boosting decision tree model. The method uses generative adversarial networks to expand an original performance database obtained from literature data and experimental data to obtain a reliable expanded performance database. The original performance database and the expanded performance database are used to form a mechanical and hole expansion performance database, and a gradient boosting decision tree model is trained and constructed. The obtained gradient boosting decision tree model can predict the mechanical properties and hole expansion properties of alloy-based metal materials. The prediction results have strong interpretability and high prediction accuracy. The method has good application prospects in the field of prediction and evaluation of the mechanical properties and hole expansion properties of alloy-based metal materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Example 1 Flow chart of the method for predicting hole expansion performance of alloy metal materials;

[0060] Figure 2 Example 2: Comparison of yield strength prediction results of low alloy high strength steel;

[0061] Figure 3 Example 2: Comparison of tensile strength prediction results of low alloy high strength steel;

[0062] Figure 4Example 2: Comparison of total elongation prediction results of low alloy high strength steel;

[0063] Figure 5 Example 2: Comparison of hole expansion rate prediction results of low alloy high strength steel. DETAILED DESCRIPTION

[0064] The word "embodiment" used herein as an "exemplary" does not necessarily mean that any embodiment described is superior or better than other embodiments. Performance index tests in the embodiments of the present invention are performed using conventional test methods in the art unless otherwise specified. It should be understood that the terms described in the embodiments of the present invention are only used to describe specific implementation methods and are not intended to limit the contents disclosed in the embodiments of the present invention.

[0065] Unless otherwise specified, the technical and scientific terms used herein have the same meanings as commonly understood by ordinary technicians in the technical field to which the embodiments of the present invention belong; other experimental methods and technical means not specifically specified in the embodiments of the present invention refer to experimental methods and technical means commonly used by ordinary technicians in the field.

[0066] The terms "substantially" and "approximately" used herein are used to describe small fluctuations. For example, they can refer to less than or equal to ±5%, such as less than or equal to ±2%, such as less than or equal to ±1%, such as less than or equal to ±0.5%, such as less than or equal to ±0.2%, such as less than or equal to ±0.1%, such as less than or equal to ±0.05%. The numerical data represented or presented in the range format herein are used only for convenience and brevity, and should therefore be flexibly interpreted as including not only the values ​​clearly listed as the limits of the range, but also all independent values ​​or sub-ranges contained in the range. For example, the numerical range of "1-5%" should be interpreted as including not only the clearly listed values ​​of 1% to 5%, but also the independent values ​​and sub-ranges within the range shown. Therefore, independent values ​​such as 2%, 3.5% and 4% and sub-ranges such as 1%-3%, 2%-4% and 3%-5% are included in this numerical range. This principle also applies to the range of only one numerical value. In addition, such an interpretation applies regardless of the width of the range or the characteristics described.

[0067] In this document, including in the claims, transitional words such as "comprises," "includes," "with," "having," "containing," "involving," "accommodating," etc. are understood to be open-ended, i.e., meaning "including but not limited to." Only the transitional words "consisting of" and "composed of" are closed transitional words.

[0068] In order to better illustrate the present invention, numerous specific details are provided in the specific examples below. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In the embodiments, some methods, means, instruments, equipment, etc. well known to those skilled in the art are not described in detail in order to highlight the gist of the present invention.

[0069] Under the premise of no conflict, the technical features disclosed in the embodiments of the present invention may be arbitrarily combined, and the resulting technical solutions belong to the contents disclosed in the embodiments of the present invention.

[0070] In some embodiments, the method for predicting the mechanical and hole expansion properties of alloy metal materials based on the generative adversarial network and the gradient boosting decision tree model comprises the steps of:

[0071] S1. Establish an original performance database of alloy metal materials; the original performance database includes the composition and content, mechanical performance parameters and hole expansion performance parameters of alloy metal materials, which are collected through two channels: literature and experimental data; usually, the composition and content of alloy metal materials refer to all metal elements that make up the alloy and the content of each metal in the alloy; generally, the mechanical and hole expansion properties of metal materials include mechanical performance parameters and hole expansion and flanging performance parameters; mechanical performance parameters include yield strength, tensile strength, uniform elongation and total elongation, and the hole expansion and flanging performance parameter is the hole expansion rate;

[0072] S2. Generate an adversarial network to expand the original performance database of alloy-based metal materials to obtain an expanded performance database; the original performance database and the expanded performance database constitute the mechanical and hole expansion performance database of alloy-based metal materials; the hole expansion performance database data obtained by the generative adversarial network expansion, including the composition and content, mechanical performance parameters and hole expansion performance parameters of alloy-based metal materials, together with the composition and content, mechanical performance parameters and hole expansion performance parameters of alloy-based metal materials obtained from literature and experimental data, constitute the mechanical and hole expansion performance database of alloy-based metal materials for training and constructing a gradient boosting decision tree model; usually, the generative adversarial network consists of a generative adversarial network. The generator and the discriminator are composed of two parts. During the training process of the generative adversarial network, the generator and the discriminator compete with each other. The generator generates more realistic data to deceive the discriminator, while the discriminator improves its discrimination ability and distinguishes the real data from the generated data more accurately. Ultimately, under ideal conditions, the generator can generate realistic data, making it difficult for the discriminator to distinguish between true and false. The data generated by the generator can be used as real sample data and together with the original performance data, form a mechanical and hole expansion performance database. The generative adversarial network effectively expands the data volume of the mechanical and hole expansion performance database, which is conducive to improving the accuracy and stability of the training gradient boosting decision tree model, and improving the prediction accuracy and generalization ability.

[0073] S3. Using a stratified sampling method, the data of the mechanical and hole expansion performance database is divided into a training set data and a test set data; usually, the data of the mechanical and hole expansion performance database is divided into two groups according to a certain ratio, one group is used as a training data set for learning and training the gradient boosting decision tree model and constructing an optimized gradient boosting decision tree model; the other group is used as a test data set for performance testing and evaluation of the constructed optimized gradient boosting decision tree model; in some embodiments, stratified sampling is performed at a ratio of 9:1 to divide the data of the performance prediction database into training set data and test set data;

[0074] S4. Use cross-validation method to divide the training set data into training data and validation data, use gradient boosting decision tree model to learn and train the training data, use validation data to correct the hyperparameters of gradient boosting decision tree model, and construct gradient boosting decision tree model;

[0075] Usually, the gradient boosting decision tree model starts from the data itself, uses the model algorithm to autonomously mine the data rules, establishes a complex mapping relationship between input variables and output variables, does not distinguish between physical mechanisms, and can make full use of the metal material mechanics and hole expansion performance data collected in the literature, the metal material mechanics and hole expansion performance data obtained by experiments, and the metal material mechanics and hole expansion performance data obtained by data expansion methods to establish a metal mechanics and hole expansion performance database as a training database for the gradient boosting decision tree model, thereby expanding the data volume of the database and improving the prediction accuracy of the gradient boosting decision tree model; and the gradient boosting decision tree model has low modeling cost and high prediction efficiency;

[0076] S5. Use the test set data to evaluate the constructed gradient boosting decision tree model to evaluate its prediction accuracy, generalization ability and prediction stability;

[0077] S6. Predicting the mechanical properties and hole expansion properties of alloy-based metal materials using the constructed gradient boosting decision tree model. Generally, the constructed gradient boosting decision tree model can be used to predict the mechanical properties and hole expansion properties of new alloy materials in the target alloy-based metal materials.

[0078] In some embodiments, the generative adversarial network is composed of a generator and a discriminator, and the method of using the generative adversarial network to obtain an expanded performance database includes:

[0079] S201, obtaining initial model parameters of the generator and the discriminator; usually the model parameters include weights, biases, etc.;

[0080] S202, using the original performance database data as the real database data of the discriminator, and generating random noise data as the input data of the generator; generally, a simple probability distribution method, such as Gaussian distribution or uniform distribution, can be used to generate random noise as the input data of the generator;

[0081] S203, training the discriminator, specifically including: extracting a batch of real sample data from the real database data, and using the generator to generate a batch of fake sample data; for example, extracting a batch of real sample data from the real data set X, and using the generator G to generate a batch of fake sample data, and calculating the loss function using the following formula:

[0082]

[0083] Among them, the loss function encourages the discriminator to classify the real data set X as real, with an output close to 1, and to classify the generated fake sample data G(z) as fake, with an output close to 0; by minimizing the loss function L of the discriminator D , update the discriminator parameters θ D ;

[0084] S204, training generator; specifically including:

[0085] Using the fake sample data G(z) generated by the generator G, the loss function of the generator is calculated using the following formula:

[0086] L G =-E z～pz(z) [logD(G(z))];

[0087] The generator minimizes the loss function so that the discriminator cannot distinguish between the generated fake sample data and the real sample data, that is, the output of D(G(z)) is close to 1, and the parameters θ of the generator are updated by minimizing the loss function of the generator. G ;

[0088] S205, repeating steps S203 and S204 multiple times; usually, the training of the discriminator and the generator are iterated alternately, and in each iteration or each round of iteration, the training repetition batch of the discriminator and the training repetition batch of the generator can be the same or different, and the specific batch depends on whether the training results meet the expected level;

[0089] S206. Adjust the hyperparameters of the generative adversarial network according to the quality of the generated fake sample data until the quality of the generated fake sample data reaches the expected level; the hyperparameters include the learning rate, the number of training times of the generator, and the number of training times of the discriminator;

[0090] S207. Generate sample data that meets expected quality using the generator of the trained generative adversarial network to obtain an expanded performance database.

[0091] In some embodiments, the method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model is disclosed, and step S4 includes:

[0092] S401. Initialize the gradient boosting decision tree model using the training data. Usually, a constant value that minimizes the loss function is selected as the initial prediction value of the gradient boosting decision tree model; for the mean square error loss function, the initial prediction value is the mean of the target value of the training data, and the calculation formula is:

[0093]

[0094] Among them, L is the loss function, yi is the target value, and F o(x) is the initial predicted value, γ is the true value;

[0095] S402, calculating the residual between the predicted value and the true value of each alloy metal material sample as the error of the current model, wherein the residual calculation formula for the mth iteration is:

[0096]

[0097] Among them, r im represents the residual of the ith sample of the alloy metal material at the mth iteration;

[0098] S403, using the calculated residual as the target value, training a new decision tree, and fitting the residual to reduce the prediction error of the gradient boosting decision tree model; wherein the decision tree of the mth iteration is expressed as:

[0099] h m (x) = Train(X, r m );

[0100] Among them, X is the training data set, r m is the residual of the alloy metal material sample of the mth iteration, h m (x) is the regression tree of round m;

[0101] S404, adding the new decision tree to the gradient boosting decision tree model, and updating the prediction value of the model; usually, the prediction result of the new regression tree is multiplied by the learning rate α to control the influence of the new regression tree on the final model. The updated gradient boosting decision tree model is:

[0102] F m (x) = F m-1 (x)+αh m (x)

[0103] Among them, F m (x) is the predicted value of the mth iteration, F m-1 (x) is the predicted value of the m-1th iteration, α is the learning rate, which ranges from 0 to 1; h m (x) is the regression tree of the mth iteration;

[0104] S405, repeat steps S402 to S404 until the predetermined number of iterations M is reached or the model performance requirements are met, and the final gradient boosting decision tree model is obtained, which is expressed as:

[0105]

[0106] Among them, F M (x) is the predicted value after the Mth iteration; F0(x) is the initial predicted value; α is the learning rate, which ranges from 0 to 1; h m (x) is the regression tree of the mth iteration; m is the number of iterations, ranging from 1 to M.

[0107] Some embodiments disclose a method for predicting the mechanical and hole expansion properties of alloy metal materials based on a generative adversarial network and a gradient boosting decision tree model, which uses a gradient boosting decision tree model to learn training data and uses grid search and cross validation to optimize parameters, wherein the cross validation is selected to be 10 fold.

[0108] In some embodiments, the prediction method for the mechanical and hole expansion properties of alloy metal materials based on the generative adversarial network and the gradient boosting decision tree model is disclosed, and the generalization ability of the gradient boosting decision tree model is evaluated using the mean square error. The calculation formula of the mean square error MSE is:

[0109]

[0110] The average relative error and determination coefficient are used to evaluate the prediction accuracy of the gradient boosting decision tree model; the calculation formula of the average relative error AARE is:

[0111]

[0112] Coefficient of determination R 2 The calculation formula is:

[0113]

[0114] The standard deviation is used to evaluate the stability of the gradient boosting decision tree model. The calculation formula of the standard deviation SD is:

[0115]

[0116] Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio.

[0117] In some embodiments, a computer-readable storage medium is disclosed, in which a computer program is stored. When the computer program is read and executed, a method for predicting the mechanical and hole expansion properties of alloy-based metal materials based on a generative adversarial network and a gradient boosting decision tree model is executed; usually, the computer program is implemented by commercial computing software such as Matlab or Python language or C language or Java language.

[0118] The technical details are further illustrated below in conjunction with embodiments.

[0119] Example 1

[0120] In Example 1, Figure 1 As shown in the figure, the prediction method of mechanical and hole expansion properties of alloy metal materials based on generative adversarial network and gradient boosting decision tree model includes:

[0121] S1. Establish an original performance database of alloy metal materials; the original performance database includes the components and contents, mechanical performance parameters and hole expansion performance parameters of alloy metal materials, which are collected through two channels: literature and experimental data;

[0122] S2. Using a generative adversarial network, the performance database of alloy-based metal materials is expanded to obtain an expanded performance database; the original performance database and the expanded performance database constitute a mechanical and hole expansion performance database of alloy-based metal materials;

[0123] S3, using stratified sampling method, the data of mechanical and hole expansion performance database is divided into training set data and test set data;

[0124] S4. Use cross-validation method to divide the training set data into training data and validation data, use gradient boosting decision tree model to learn and train the training data, use validation data to correct the hyperparameters of gradient boosting decision tree model, and construct gradient boosting decision tree model;

[0125] S5. Use the test set data to evaluate the constructed gradient boosting decision tree model to evaluate its prediction accuracy, generalization ability and prediction stability;

[0126] S6. Use the constructed gradient boosting decision tree model to predict the mechanical properties and hole expansion performance of alloy metal materials.

[0127] Example 2

[0128] In Example 2, the method for predicting the mechanical and hole expansion properties of alloy metal materials based on the generative adversarial network and the gradient boosting decision tree model includes:

[0129] S1. Establish an original performance database of alloy metal materials; the original performance database includes the components and contents of alloy metal materials, mechanical performance parameters and hole expansion performance parameters, which are collected through two channels: literature and experimental data; among them, the alloy metal materials are low-alloy high-strength steels, and the data set of the original performance database has several input features and 4 output variables; the input features are independent variables used by the model to predict the target, including the mass fraction of the chemical composition of the material and the manufacturing process parameters; among them, the chemical composition includes several elements such as C, Si, Mn, Cr, Mo, Ti, Nb, V, N, P, S, Alt, O, Ni, B, etc., and the manufacturing process parameters include several manufacturing processes such as solution temperature, rolling start temperature, rolling end temperature, first air cooling temperature, first water cooling temperature, second air cooling temperature, second water cooling temperature, third air cooling temperature, coiling temperature, annealing temperature, aging temperature, etc.; the output variable is the target to be predicted by the model, which is the result that the model attempts to output after learning the input features. The four output variables are yield strength, tensile strength, total elongation and hole expansion rate;

[0130] S2. Generate adversarial networks to expand the performance database of alloy metal materials to obtain an expanded performance database. The original performance database and the expanded performance database constitute the mechanical and hole expansion performance database of alloy metal materials. Since the experimental environment parameters have a great influence on the output variables, expanding the experimental data can reduce the influence of the experimental environment parameters on the output variables. Generate adversarial network processing is performed on 19 experimental data, and the amount of experimental data is expanded to 95. Specifically, it includes:

[0131] S201, obtaining initial model parameters of the generator and the discriminator; usually the model parameters include weights, biases, etc.;

[0132] S202, using the original performance database data as the real database data of the discriminator, and using the Gaussian distribution method to generate random noise data as input data of the generator;

[0133] S203, training the discriminator, specifically including: extracting a batch of real sample data from the real database data, and using the generator to generate a batch of fake sample data; for example, extracting a batch of real sample data from the real data set X, and using the generator G to generate a batch of fake sample data, and calculating the loss function using the following formula:

[0134]

[0135] Among them, the loss function encourages the discriminator to classify the real data set X as real, with an output close to 1, and to classify the generated fake sample data G(z) as fake, with an output close to 0; by minimizing the loss function L of the discriminator D , update the discriminator parameters θ D ;

[0136] S204, training generator; specifically including:

[0137] Using the fake sample data G(z) generated by the generator G, the loss function of the generator is calculated using the following formula:

[0138] L G =-E z～pz(z) [logD(G(z))];

[0139] The generator minimizes the loss function so that the discriminator cannot distinguish between the generated fake sample data and the real sample data, that is, the output of D(G(z)) is close to 1, and the parameters θ of the generator are updated by minimizing the loss function of the generator. G ;

[0140] S205, repeat steps S203 and S204 multiple times to reach the required data volume;

[0141] S206. Adjust the hyperparameters of the generative adversarial network according to the quality of the generated fake sample data until the quality of the generated fake sample data reaches the expected level; the hyperparameters include the learning rate, the number of training times of the generator, and the number of training times of the discriminator;

[0142] S207, using the trained generator of the generative adversarial network to generate sample data that meets the expected quality, and obtaining an expanded performance database;

[0143] S3. Using the stratified sampling method, the data of the mechanical and hole expansion performance database are divided into training set data and test set data in a ratio of 9:1;

[0144] S4. Use cross-validation to divide the training set data into training data and validation data, use the gradient boosting decision tree model to learn and train the training data, use the validation data to correct the hyperparameters of the gradient boosting decision tree model, and build a gradient boosting decision tree model; use grid search and cross-validation to optimize the parameters, and select 10 folds for cross-validation; use grid search and cross-validation to optimize the model parameters, so as to find the best hyperparameter combination for model performance by adjusting the hyperparameters of the model, so that the model has the best generalization ability on unknown data; usually, grid search is used to tune hyperparameters in machine learning, by listing all possible parameter combinations in a given hyperparameter space, and evaluating the model performance of each combination one by one, and finally selecting the best hyperparameter combination; the unbiasedness of the model means that the performance evaluation of the model can reflect its true generalization ability without systematic bias. Specifically include:

[0145] S401, using the training data to initialize the gradient boosting decision tree model; usually a constant value that minimizes the loss function is selected as the initial prediction value of the gradient boosting decision tree model; for the mean square error loss function, the initial prediction value is the mean of the target value of the training data, and the calculation formula is:

[0146]

[0147] Among them, L is the loss function, yi is the target value, and F o(x) is the initial predicted value, γ is the true value;

[0148] S402, calculating the residual between the predicted value and the true value of each alloy metal material sample as the error of the current model, wherein the residual calculation formula for the mth iteration is:

[0149]

[0150] Among them, r im represents the residual of the ith sample of the alloy metal material at the mth iteration;

[0151] S403, using the calculated residual as the target value, training a new decision tree, and fitting the residual to reduce the prediction error of the gradient boosting decision tree model; wherein the decision tree of the mth iteration is expressed as:

[0152] h m (x) = Train(X, r m );

[0153] Among them, X is the training data set, r m is the residual of the alloy metal material sample of the mth iteration, h m (x) is the regression tree of round m;

[0154] S404, adding the new decision tree to the gradient boosting decision tree model, and updating the prediction value of the model; usually, the prediction result of the new regression tree is multiplied by the learning rate α to control the influence of the new regression tree on the final model. The updated gradient boosting decision tree model is:

[0155] F m (x) = F m-1 (x)+αh m (x)

[0156] Among them, F m (x) is the mth predicted value, F m-1 (x) is the m-1th prediction value, α is the learning rate, which ranges from 0 to 1; h m (x) is the mth regression tree;

[0157] S405, repeat steps S402 to S404 until the predetermined number of iterations M is reached, where M is 500, or the model performance requirements are met, and the final gradient boosting decision tree model is obtained, which is expressed as:

[0158]

[0159] Among them, F M (x) is the predicted value after the Mth iteration; F0(x) is the initial predicted value; α is the learning rate, which ranges from 0 to 1; h m (x) is the regression tree of the mth iteration; m is the number of iterations, ranging from 1 to M;

[0160] S5. Use the test set data to evaluate the constructed gradient boosting decision tree model to evaluate its prediction accuracy, generalization ability and prediction stability;

[0161] S6. Use the constructed gradient boosting decision tree model to predict the mechanical properties and hole expansion properties of alloy metal materials;

[0162] Example 2 is implemented in Python; the kernel function selected is a Gaussian kernel function, which includes a model parameter gamma; the penalty coefficient C is pre-specified in the range of 1×10 -2 ~1×10 6 , the support distance e ranges from 0 to 0.6, and the Gamma ranges from 0.01 to 0.7;

[0163] The prediction accuracy, generalization ability and stability of the model are evaluated using the hole expansion performance data set of the test set. The trained model is used to predict the test set data to obtain the MSE, AARE, SD and R on the test set. 2 .

[0164] The trained GBDT model predicts the test data, and the prediction results show that:

[0165] Figure 2 A comparison chart of the predicted yield strength of low alloy high strength steel using the GBDT model is given. As can be seen from the figure, the data points are basically close to the 45° dashed line, indicating that the predicted value of the yield strength is very close to the true value; usually, the true value is obtained through experimental data as the data in the training set; the determination coefficient of the prediction result is 0.805, indicating that the constructed yield strength prediction model has a high prediction accuracy.

[0166] Figure 3A comparison chart of the tensile strength prediction of low alloy high strength steel by GBDT model is given. As can be seen from the figure, the data points are basically close to the 45° dashed line, and the determination coefficient of the prediction result is 0.944; the GBDT model achieves a high tensile strength prediction accuracy, and the accuracy of the prediction results of the training set and the test set shows good consistency.

[0167] Figure 4 A comparison chart of the total elongation predicted by the GBDT model for low-alloy high-strength steel is given. Due to the low total elongation value and large experimental error, the predicted value of elongation by machine learning is often lower than the predicted value of strength; as can be seen from the figure, the data points are basically close to the 45° dotted line, and the determination coefficient of the prediction result is 0.749. The GBDT model achieves a high total elongation prediction accuracy, and the accuracy of the prediction results of the training set and the test set shows good consistency.

[0168] Figure 5 A comparison chart of the hole expansion rate prediction of low alloy high strength steel using the GBDT model is given. Since the experimental error of the hole expansion rate is large, the error bar often exceeds 15%, so the prediction of the hole expansion rate is difficult; as can be seen from the figure, most of the data points are close to the 45° dashed line, and the determination coefficient of the prediction result is 0.787, indicating that the constructed hole expansion rate prediction model has a high prediction accuracy.

[0169] The embodiment of the present invention discloses a method for predicting the mechanical and hole expansion properties of alloy-based metal materials based on a generative adversarial network and a gradient boosting decision tree model. The method uses generative adversarial networks to expand an original performance database obtained from literature data and experimental data to obtain a reliable expanded performance database. The original performance database and the expanded performance database are used to form a mechanical and hole expansion performance database, and a gradient boosting decision tree model is trained and constructed. The obtained gradient boosting decision tree model can predict the mechanical properties and hole expansion properties of alloy-based metal materials. The prediction results have strong interpretability and high prediction accuracy. The method has good application prospects in the field of prediction and evaluation of the mechanical properties and hole expansion properties of alloy-based metal materials.

[0170] The technical solutions disclosed in the embodiments of the present invention and the technical details disclosed in the embodiments are merely illustrative of the inventive concept of the present invention and do not constitute a limitation on the technical solutions of the embodiments of the present invention. Any conventional changes, replacements or combinations of the technical details disclosed in the embodiments of the present invention have the same inventive concept as the present invention and are within the protection scope of the claims of the present invention.

Claims

1. A method for predicting the mechanical and hole expansion properties of alloy metal materials based on generative adversarial networks and gradient boosting decision tree models, characterized in that: Includes steps: S1. Establishing an original performance database of alloy metal materials; the original performance database includes components and contents, mechanical performance parameters and hole expansion performance parameters of alloy metal materials, which are collected through two channels: literature and experimental data; S2. Using a generative adversarial network, the original performance database of alloy-based metal materials is expanded to obtain an expanded performance database; the original performance database and the expanded performance database constitute a mechanical and hole expansion performance database of alloy-based metal materials; S3, using stratified sampling method, the data of mechanical and hole expansion performance database is divided into training set data and test set data; S4, using the cross-validation method to divide the training set data into training data and validation data, using the gradient boosting decision tree model to learn and train the training data, using the validation data to calibrate the hyperparameters of the gradient boosting decision tree model, and constructing the gradient boosting decision tree model; S5. Use the test set data to evaluate the constructed gradient boosting decision tree model to evaluate its prediction accuracy, generalization ability and prediction stability; S6. Use the constructed gradient boosting decision tree model to predict the mechanical properties and hole expansion performance of alloy metal materials.

2. The method according to claim 1, characterized in that: The generative adversarial network is composed of a generator and a discriminator. The method of obtaining an expanded performance database using the generative adversarial network includes: S201, obtaining initial model parameters of the generator and the discriminator; S202, using the original performance database data as the real database data of the discriminator, and generating random noise data as the input data of the generator; S203, training the discriminator, specifically including: extracting a batch of real sample data from the real database data, using the generator to generate a batch of fake sample data, and calculating the loss function using the following formula: Among them, the loss function encourages the discriminator to classify the real sample data as real and the generated fake sample data as fake, and updates the parameters of the discriminator by minimizing the loss function of the discriminator; S204, training generator; specifically including: Using the fake sample data generated by the generator, the loss function of the generator is calculated using the following formula: L G =-E z～pz(z) [logD(G(z))]; The generator minimizes the loss function so that the discriminator cannot distinguish between the generated fake sample data and the real sample data, and updates the parameters of the generator by minimizing the loss function of the generator; S205, repeat steps S203 and S204 multiple times; S206. Adjust the hyperparameters of the generative adversarial network according to the quality of the generated fake sample data until the quality of the generated fake sample data reaches the expected level; the hyperparameters include a learning rate, a number of training times of the generator, and a number of training times of the discriminator; S207. Generate sample data that meets expected quality using the generator of the trained generative adversarial network to obtain an expanded performance database.

3. The method according to claim 1, characterized in that: In step S3, stratified sampling is performed according to a ratio of 9:1 using a stratified sampling method to divide the data of the performance prediction database into training set data and test set data.

4. The method according to claim 4, characterized in that: Step S4 includes: S401, using training data to initialize the gradient boosting decision tree model, for the mean square error loss function, the initial prediction value is the mean of the target value of the training data, and the calculation formula is: Among them, L is the loss function, yi is the target value, and F o(x) is the initial predicted value, γ is the true value; S402, calculating the residual between the predicted value and the true value of each alloy metal material sample as the error of the current gradient boosting decision tree model, wherein the residual calculation formula for the mth iteration is: Among them, r im represents the residual of the ith alloy-based metal material sample of the mth iteration; S403, taking the calculated residual as the target value, training a new decision tree, and fitting the residual; wherein the decision tree of the mth iteration is expressed as: h m (x)=Train(X,r m ); Among them, X is the training data set, r m is the residual of the current alloy metal material sample at the mth iteration, h m (x) is the regression tree of round m; S404, adding the new decision tree to the gradient boosting decision tree model, and updating the prediction value of the model; the updated gradient boosting decision tree model is: F m (x)=F m-1 (x)+αh m (x) Among them, F m (x) is the predicted value of the mth iteration, F m-1 (x) is the predicted value of the m-1th iteration, α is the learning rate, which ranges from 0 to 1; h m (x) is the regression tree of the mth iteration; S405, repeat steps S402 to S404 until the predetermined number of iterations M is reached or the model performance requirements are met, and the final gradient boosting decision tree model is obtained, which is expressed as: Among them, F M (x) is the predicted value after the Mth iteration; F0(x) is the initial predicted value; α is the learning rate, which ranges from 0 to 1; h m (x) is the mth regression tree; m is the number of iterations, ranging from 1 to M.

5. The method according to claim 4, characterized in that The gradient boosting decision tree model is used to learn the training data, and grid search and cross validation are used to optimize the parameters, where the cross validation is selected as 10 folds.

6. The method according to claim 1, characterized in that The mean square error (MSE) is used to evaluate the generalization ability of the gradient boosting decision tree model. The calculation formula is: Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio.

7. The method according to claim 1, characterized in that The prediction accuracy of the gradient boosting decision tree model was evaluated using mean relative error and coefficient of determination. The calculation formula of average relative error AARE is: Coefficient of determination R 2 The calculation formula is: Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio.

8. The method according to claim 1, characterized in that: The standard deviation SD is used to evaluate the stability of the gradient boosting decision tree model. The calculation formula is: Among them, n is the number of test set data, N i_pre To predict the hole expansion rate, N i_exp is the test hole expansion ratio, and AARE is the average relative error.

Images